Automated written corrective feedback: Error-correction performance and timing of delivery

Abstract

To the extent automated written corrective feedback (AWCF) tools such as Grammarly are based on sophisticated error-correction technologies, such as machine-learning techniques, they have the potential to find and correct more common L2 error types than simpler spelling and grammar checkers such as the one included in Microsoft Word (technically known as MS-NLP). Moreover, AWCF tools can deliver feedback synchronously, although not instantaneously, as often appears to be the case with MS-NLP. Cognitive theory and recent L2 research suggest that synchronous corrective feedback may aid L2 development, but also that error-flagging at suboptimal times could cause disfluencies in L2 students’ writing processes. To contribute to the knowledge needed for appropriate application of this new genre of writing-support technology, we evaluated Grammarly’s capacity to address common L2 problem areas, as well as issues with its feedback-delivery timing, using MS-NLP as a benchmark. Grammarly was found to flag 10 times as many common L2 error types as MS-NLP in the same corpus of student texts while also displaying an average 17.5-second delay in feedback delivery, exceeding the distraction-potential threshold defined for the L2 student writers in our sample. Implications for the use of AWCF tools in L2 settings are discussed.